Refa: remove temperature since some LLMs fail to support. (#8981)

### What problem does this PR solve?


### Type of change

- [x] Refactoring

graphrag/entity_resolution.py

@@ -152,7 +152,6 @@ class EntityResolution(Extractor):
         )
 
     async def _resolve_candidate(self, candidate_resolution_i: tuple[str, list[tuple[str, str]]], resolution_result: set[str], resolution_result_lock: trio.Lock):
-        gen_conf = {"temperature": 0.5}
         pair_txt = [
             f'When determining whether two {candidate_resolution_i[0]}s are the same, you should only focus on critical properties and overlook noisy factors.\n']
         for index, candidate in enumerate(candidate_resolution_i[1]):
@@ -171,7 +170,7 @@ class EntityResolution(Extractor):
         async with chat_limiter:
             try:
                 with trio.move_on_after(120) as cancel_scope:
-                    response = await trio.to_thread.run_sync(self._chat, text, [{"role": "user", "content": "Output:"}], gen_conf)
+                    response = await trio.to_thread.run_sync(self._chat, text, [{"role": "user", "content": "Output:"}], {})
                 if cancel_scope.cancelled_caught:
                     logging.warning("_resolve_candidate._chat timeout, skipping...")
                     return

graphrag/general/community_reports_extractor.py

@@ -90,11 +90,10 @@ class CommunityReportsExtractor(Extractor):
                 "relation_df": rela_df.to_csv(index_label="id")
             }
             text = perform_variable_replacements(self._extraction_prompt, variables=prompt_variables)
-            gen_conf = {"temperature": 0.3}
             async with chat_limiter:
                 try:
                     with trio.move_on_after(80) as cancel_scope:
-                        response = await trio.to_thread.run_sync( self._chat, text, [{"role": "user", "content": "Output:"}], gen_conf)
+                        response = await trio.to_thread.run_sync( self._chat, text, [{"role": "user", "content": "Output:"}], {})
                     if cancel_scope.cancelled_caught:
                         logging.warning("extract_community_report._chat timeout, skipping...")
                         return

graphrag/general/graph_extractor.py

@@ -105,10 +105,9 @@ class GraphExtractor(Extractor):
             **self._prompt_variables,
             self._input_text_key: content,
         }
-        gen_conf = {"temperature": 0.3}
         hint_prompt = perform_variable_replacements(self._extraction_prompt, variables=variables)
         async with chat_limiter:
-            response = await trio.to_thread.run_sync(lambda: self._chat(hint_prompt, [{"role": "user", "content": "Output:"}], gen_conf))
+            response = await trio.to_thread.run_sync(lambda: self._chat(hint_prompt, [{"role": "user", "content": "Output:"}], {}))
         token_count += num_tokens_from_string(hint_prompt + response)
 
         results = response or ""
@@ -118,7 +117,7 @@ class GraphExtractor(Extractor):
         for i in range(self._max_gleanings):
             history.append({"role": "user", "content": CONTINUE_PROMPT})
             async with chat_limiter:
-                response = await trio.to_thread.run_sync(lambda: self._chat("", history, gen_conf))
+                response = await trio.to_thread.run_sync(lambda: self._chat("", history, {}))
             token_count += num_tokens_from_string("\n".join([m["content"] for m in history]) + response)
             results += response or ""
 

graphrag/general/mind_map_extractor.py

@@ -171,9 +171,8 @@ class MindMapExtractor(Extractor):
             self._input_text_key: text,
         }
         text = perform_variable_replacements(self._mind_map_prompt, variables=variables)
-        gen_conf = {"temperature": 0.5}
         async with chat_limiter:
-            response = await trio.to_thread.run_sync(lambda: self._chat(text, [{"role": "user", "content": "Output:"}], gen_conf))
+            response = await trio.to_thread.run_sync(lambda: self._chat(text, [{"role": "user", "content": "Output:"}], {}))
         response = re.sub(r"```[^\n]*", "", response)
         logging.debug(response)
         logging.debug(self._todict(markdown_to_json.dictify(response)))

graphrag/search.py

@@ -45,7 +45,7 @@ class KGSearch(Dealer):
         ty2ents = trio.run(lambda: get_entity_type2sampels(idxnms, kb_ids))
         hint_prompt = PROMPTS["minirag_query2kwd"].format(query=question,
                                                           TYPE_POOL=json.dumps(ty2ents, ensure_ascii=False, indent=2))
-        result = self._chat(llm, hint_prompt, [{"role": "user", "content": "Output:"}], {"temperature": .5})
+        result = self._chat(llm, hint_prompt, [{"role": "user", "content": "Output:"}], {})
         try:
             keywords_data = json_repair.loads(result)
             type_keywords = keywords_data.get("answer_type_keywords", [])